Java 7, bukkit, and the experimental GC

So, when I run my server like this:

Code:

java -server -Xms8192M -Xmx8192M -XX:+UseConcMarkSweepGC -XX:+UseParNewGC -XX:+CMSIncrementalPacing -XX:parallelGCThreads=6 -XX:+AggressiveOpts -jar $SERVER_FILE nogui

It works fine, but if I run it like this:

Code:

java -server -Xms8192M -Xmx8192M -XX:+UnlockExperimentalVMOptions -XX:+UseG1GC -jar $SERVER_FILE nogui

I get crap like this:

Code:

13:03:55 [INFO] Loaded world: world
13:03:55 [SEVERE] java.lang.NullPointerException
13:03:55 [SEVERE]       at net.minecraft.server.World.h(World.java:1468)
13:03:55 [SEVERE]       at net.minecraft.server.World.g(World.java:1434)
13:03:55 [SEVERE]       at net.minecraft.server.MinecraftServer.h(MinecraftServer.java:348)
13:03:55 [SEVERE]       at net.minecraft.server.MinecraftServer.run(MinecraftServer.java:272)
13:03:55 [SEVERE]       at net.minecraft.server.ThreadServerApplication.run(SourceFile:366)
13:03:55 [SEVERE] Unexpected exception
java.lang.NullPointerException
        at net.minecraft.server.World.h(World.java:1468)
        at net.minecraft.server.World.g(World.java:1434)
        at net.minecraft.server.MinecraftServer.h(MinecraftServer.java:348)
        at net.minecraft.server.MinecraftServer.run(MinecraftServer.java:272)
        at net.minecraft.server.ThreadServerApplication.run(SourceFile:366)

http://pastie.org/1667332

if I try to make a new world, it generates one, then as soon as someone comes in, errors with ^ error, and when the server is restarted generates again.

Any idea why the experimental GC would cause that?

 

Are you using tmpfs?

 

Yea, server runs on a ramdisk, why? I haven't touched any swap as of yet.

 

I have been experimenting with JVM/GC settings for months now. And recently I have noticed the same issue you just posted in this combination only:

- tmpfs
- Java 7
- G1GC

Not using tmpfs solves the issue, so does using Java 7 without G1GC. Very odd, but G1GC does some very fancy things, and it's far from the first application-level issue I have seen with it..

 

Ah, darn, I was hoping to get some lower mem usage with mc :/
Not using my ramdisk isn't really an option

 

Well then I suggest dropping G1GC, for us, G1GC is actually giving less performance with MC 1.3, even though it did better with 1.2

 

Aye, the first line is what I'm using anyway (like I said, the second doesn't work at all).
If you've got any tips on what I could add/remove to the first string then feel free :3
[MERGETIME="1300077821"][/MERGETIME]
Heh, it doesn't like a standard disk either. Oh well, abandoning that.

 

It worked fine in 1.2, but I also used it for a while in 1.3, so I am not sure what's happening. It could be that any 1.3 map which has been used by non-G1GC, can no longer use G1GC, how ridiculous it may sound <_<

Hmm did you test on 132+ ?

Seems 131 does not have this issue

EDIT by Moderator: merged posts, please use the edit button instead of double posting.

 

Say again? 131? 132?

 

been tryin g1gc with java7, pauses are extremely fast, however it behaves oddly, not sure if i am going to stay with it